Search

Accurate four-hour-ahead PV power prediction is crucial to the utilization of PV power. Conventional methods focus on using historical data directly. This paper addresses this issue from a new perspective of Numerical Weather Prediction (NWP) optimization. This paper refers to the predicted PV power given by NWP minus the actual PV power as PV NWP error, analyzes the temporal statistical characteristics of PV NWP error time series through the Ljung-Box test and autocorrelation function, reveals the significant temporal autocorrelation in PV NWP error and verifies the statistical predictability of PV NWP error for the first time. Meanwhile, this paper introduces transfer entropy to verify that introducing multiple meteorological variables including temperature, solar irradiation, clear-sky solar irradiation, cloud thickness, etc. can help predict future PV NWP error better for the first time. Based on the fusion of multiple meteorological variables and artificial neural networks, this paper proposes multiple meteorological variables-aided PV NWP error correction to realize more accurate four-hour-ahead PV power prediction for the first time. Kernel density estimation is used to obtain the probabilistic forecast results. Through experiments in three-year actual data of Brussels, the superiorities of the proposed method and the significant improvement over conventional methods are verified. In comparison to the original NWP method, the proposed approach demonstrates significant improvements in accuracy. Specifically, it achieves a reduction in mean absolute error ranging from 25.04% to 48.12% for 1–4 step predictions, 14.80% to 21.27% for 5–8 step predictions, 6.40% to 11.10% for 9–12 step predictions, and 2.18% to 4.45% for 13–16 step predictions. [ABSTRACT FROM AUTHOR]

Accurate photovoltaic (PV) power forecast is crucial for carbon neutrality. Current researches on PV power forecast mainly focus on using temporal information from single PV station, and the spatial information in multiple PV power stations are often neglected. To address this problem, this paper introduces Moran index to verify the spatial autocorrelation of PV power for the first time, uses Granger causality test and transfer entropy to reveal the spatial information gain in PV power forecast for the first time, and proposes a novel spatial–temporal probabilistic PV forecast method using deep Graph Gated Recurrent Unit (GraphGRU) network-based spatial–temporal multi-task learning and Kernel Density Estimation (KDE). Deep GraphGRU combines the advantages of Graph Convolutional Network (GCN) in spatial feature extraction and the advantages of Gated Recurrent Unit (GRU) network in temporal feature extraction, and thus has strong ability to extract spatial–temporal information in historical data of multiple different PV power stations. Through GraphGRU, temporal dependency information extracted from historical data of multiple PV stations can promote each other to improve the forecast accuracy of each PV stations. KDE is used for estimating the joint probabilistic density function and giving the spatial–temporal probabilistic confidence interval of PV power. Experiments were performed in the five-year actual PV power data from 11 provinces of Belgium and the three-year solar irradiation data from 12 places in China to verify the superiorities of the proposed method. Comparison with conventional spatial–temporal and temporal forecast methods show that the proposed GraphGRU-based spatial–temporal forecast method can extract the spatial–temporal information from multiple PV power stations well and significantly outperform conventional temporal forecast methods. [ABSTRACT FROM AUTHOR]

Accurate carbon pricing guidance is of great importance for the inhibition of excessive carbon dioxide emissions. Aiming at improving forecast performance, a number of carbon price forecasting models have been proposed based on the combination or multiscale hybrid frameworks. However, most of these hybrid models cannot easily cast a perfect reflection of erratic fluctuation in carbon trading schemes due to lack of judgment on the trend or inaccurate trend reconstruction. In this study, a novel filter-based modeling with Hodrick–Prescott (HP) filter, that can identify repeated up and down structural features around a certain carbon price, negotiates the obstacle of the parallel–series hybridization concerning the linear and the nonlinear model identification. The residual decomposition scheme with adaptive noise is carried out on the random and nonlinear component for error correction to filter-based models. Moreover, Bayesian optimization adjusts the structure of seven single-hidden layer feedforward neural networks (SLFNs) and the inputs to provide the best generalization performance. The proposed filter-hybrid model using kernel extreme learning machine as the final nonlinear integrator has better stability to the parameters, and has the superiority over the parallel–series and allocation-based models from a statistical perspective. Comparing with existing data-driven models, our proposed model is competitive in view of prediction accuracy and time cost in the majority of carbon futures trading cases. • A new hybrid ARIMA–SLFNs method is proposed for carbon price forecasting. • The introduction of HP filter avoids making strong assumptions of single model. • The filter model's performance is further improved by residual decomposition. • The hyperparameter of the proposed hybrid framework is optimized by BOA. • The model has advantages compared with analogous models and relevant studies. [ABSTRACT FROM AUTHOR]

Recent research has shown that introducing online data can significantly improve forecasting ability. This study considers several popular single-model machine learning methods and a stacking multiple-model ensemble learning strategy. These are used with online data from Google Trends to forecast crude oil prices. The study first selects dozens of alternative Google Trends, which may capture crude oil price fluctuations. A co-integration test and Granger causality analysis are used to investigate the effect of Google Trends on crude oil prices. Then, the multiple-model methods are compared with several popular single-model machine learning methods that are used to forecast crude oil prices. These methods are used with Google Trends that have a significant relationship with the crude oil price. Experimental results indicate that introducing Google Trends can improve the forecasting performance; multiple-model methods also outperform several popular single-model machine learning methods in terms of prediction accuracy. • We develop an effective framework of oil price forecasting with Google trends. • The framework includes skillful indicators selection and forecasting methods. • Practical study finds Google trends that significantly impact on crude oil price. • The proposed multiple-model methods show obvious advantages in practical study. [ABSTRACT FROM AUTHOR]

Precise forecast of strong nonlinear behaviors of magnetorheological energy absorber (MREA) plays a significant role in its semi-active control and engineering applications. This study proposes a hybrid model based on the support vector regression (SVR) and adaptive differential evolution (i.e., JADE) algorithm for MREA, called the JADE-SVR model. The JADE algorithm is employed to find the optimal hyperparameters of SVR to improve prediction accuracy and generalization ability. The JADE-SVR model is developed using the training data randomly partitioned from an impact test database, and then applied to predict the nonlinear behaviors of MREA in the test set, including force–time, force–displacement and force–velocity characteristics. It is demonstrated that the JADE-SVR model is capable of forecasting strong nonlinear behaviors of MREA under various impact conditions. Moreover, the JADE-SVR model is compared with other SVR-based models to validate its prediction accuracy and convergence rate further. In addition to four statistical indicators, a novel desirability function is used to assess the overall performance of these models. The results show that the JADE-SVR model has the optimal values for all the performance indices, and the largest overall desirability of 0.6274. Therefore, it can be concluded that the JADE-SVR model is the best one for forecasting the dynamic behaviors of MREA under high-impact loadings. [Display omitted] • A new SVR-based model is proposed for forecasting strong nonlinear behaviors of MREA. • The hyperparameters of SVR are optimized by using the adaptive differential evolution algorithm. • A comprehensive comparison of the four SVR-based models is conducted using a novel desirability function. • The proposed JADE-SVR model has the best prediction performance compared with other models. [ABSTRACT FROM AUTHOR]